Check for nonlocal goto in check_maybe_invariant

Check for nonlocal goto in check_maybe_invariant

[Zdenek Dvorak]

Hello,

this code is in loop.c:loop_invariant_p:

    case REG:

      if ((x == frame_pointer_rtx || x == hard_frame_pointer_rtx
           || x == arg_pointer_rtx || x == pic_offset_table_rtx)
          && ! current_function_has_nonlocal_goto)
        return 1;

Why do we check current_function_has_nonlocal_goto here?

Zdenek

Re: Check for nonlocal goto in check_maybe_invariant

[Jim Wilson]

Zdenek Dvorak wrote:
> this code is in loop.c:loop_invariant_p:
> Why do we check current_function_has_nonlocal_goto here?

If you have a nonlocal goto inside a loop, then you are going to have 
insns that modify the frame pointer inside the loop.  Of course, you are 
going to immediately exit the loop right after changing the fp, but 
maybe there was a case where part of the non-local goto sequence was 
incorrectly optimized because we thought the fp was loop invariant.

The check was added here:
Sun Apr 13 08:15:31 1997  Bernd Schmidt 
<crux@Pool.Informatik.RWTH-Aachen.DE>

         * loop.c (invariant_p, case REG): Pointers into frame are not
         invariants if function has nonlocal gotos.
         * reload1.c (reload): If function has nonlocal label, mark all
         caller-saved regs as used.

This predates gcc.gnu.org, so I can't easily find a mail message 
discussing it.
-- 
Jim Wilson, GNU Tools Support, http://www.SpecifixInc.com